Generate bounded ideas
Start from simple, inspectable hypotheses instead of an AI story that cannot be reproduced.

Give RLXBT market data. It generates testable strategies, includes trading costs, and tries to break every result out of sample — before you risk money on it.
Real BTC data included · Result in about 20 seconds · No credit card
Real evidence verdict
BTCUSDT · 1h · 4,999 bars
Hypotheses tested
12
Rejected
10
OOS windows positive
67%
Monte Carlo p5
−23.97%
Why it failed
The best in-sample result degraded out of sample. Monte Carlo downside remained too large to promote it.
A rejected strategy is a useful result: it is one less false edge to deploy.
One job
RLXBT does not promise profitable signals. It gives your idea a reproducible test and shows exactly where the evidence holds or breaks.
Start from simple, inspectable hypotheses instead of an AI story that cannot be reproduced.
Run the event-driven Rust engine with fees, slippage, realistic fills and enough trades to matter.
Use out-of-sample windows, Monte Carlo and sensitivity tests to expose fragile or overfit results.
Same engine · your server
Deploy the hardened RLXBT image with Docker Compose on Ubuntu x86_64 or ARM64. Enter your license key, connect your agent to the private endpoint, and keep researching without a desktop session.
$ docker compose up -d
✓ license activated for this server
✓ engine healthy on 127.0.0.1:8142
agent → backtest → reject overfit
agent → walk_forward → evidence report
Choose by outcome
Explore
Test the workflow on your data and discover candidates.
Prove
Attack overfitting before you deploy a strategy.
Questions
Start with a real BTC dataset. No account, no configuration, and no promise that the answer will be comfortable.